Foundations of Geometric Deep Learning
Abstract
Geometric deep learning aims to generalize neural models to non-Euclidean domains such as graphs and manifolds. The field has made promising advances and remarkable performance improvements, especially in studying social networks, recommendation systems, drug discovery, anomaly detection, and urban intelligence. In this project, we develop a foundational understanding of geometric deep learning, its capabilities, limitations, and applications. We build on and advance the theory of graph limits to study the robustness, transferability, and scalability of Graph Neural Network (GNN) learning models. In particular, we will take advantage of the powerful analytical and algorithmic toolkit developed for graphons to analyze the performance of graph neural networks on graph-structured data.
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Feb 29, 2024
- Source ID
- FA95502310251
Entities
People
- Amin Saberi
Organizations
- Air Force Office of Scientific Research
- Stanford University
- United States Air Force